Daniel Grenemark (@desek) header
Daniel Grenemark (@desek) profile picture

Burning tokens for a better world

Stockholm, Sweden
Joined April 2009
Photo by Daniel Grenemark (@desek)Photo by Daniel Grenemark (@desek)Photo by Daniel Grenemark (@desek)Photo by Daniel Grenemark (@desek)Photo by Daniel Grenemark (@desek)Photo by Daniel Grenemark (@desek)Photo by Daniel Grenemark (@desek)Photo by Daniel Grenemark (@desek)Photo by Daniel Grenemark (@desek)Photo by Daniel Grenemark (@desek)
Daniel Grenemark retweeted
πŸŒŠπŸ’™ WE WERE ALWAYS BLUE
935
3,676
1,140
28,568
7,644,508
Daniel Grenemark retweeted
JUST IN: Uber and Ponyai plan to expand autonomous rides to β€œmore European cities.”
3
4
1
33
6,394
Daniel Grenemark retweeted
We ran DeepSeek V4 Flash through 4 more agent harnesses (Hermes Agent, Pi Agent, Prime Agent, Deep Agents) on 30 challenging agentic tasks. Pi Agent was the cheapest harness and passed the most tasks 🧡🧡
187
216
155
2,340
984,590
Daniel Grenemark retweeted
FULL EPISODE 00:02:45 - What is Pi 00:08:54 - Personal Development Environment 00:18:59 - Context Sensitive 00:21:14 - String Builders 00:25:53 - Why build your own pi 00:40:21 - Time Between Prompts 00:43:00 - Models and Burnout 01:02:08 - How to get started with Pi
4
11
2
153
29,897
Why are you not running your agent overnight? The machine is on, the work still needs to get done, and the budget is the same budget. Eight hours buys one goal, one branch, and a report waiting at 08:00. Tell me what you would point it at first. I run one goal per night: /goal improve {target} in {scope} from {baseline} toward {goal state}, without {constraint}, until tomorrow 08:00. Then report where we started, what moved and by how much, why it moved, where we stand now, and what you recommend next.
34
I can now see what every agent session costs, which repo it spent it on, and what it did with it. Claude Code and pi, same dashboard, my laptop. Before this the only signal was a card statement a month later with one number on it. Shipped a local-first stack that closes it: β†’ one πšŒπš˜πš–πš™πš˜πšœπšŽ πšžπš™ from a fresh clone, one loopback port, nothing leaves the machine β†’ cost, tokens, sessions, tool decisions, lines of code, commits, PRs on a provisioned dashboard β†’ every session as a readable conversation, each turn and tool call carrying its own tokens, cost, latency β†’ read-only MCP: 22 tools, 0 write tools, so the agent queries its own history Claude Code emits OpenTelemetry natively. pi did not, so I shipped the extension too: πš™πš’ πš’πš—πšœπšπšŠπš•πš• πš—πš™πš–:@πšπšŽπšœπšŽπš”/πš™πš’-πš˜πš™πšŽπš—πšπšŽπš•πšŽπš–πšŽπšπš›πš’ Every instrument carries Claude Code's own name with the prefix swapped, πšŒπš•πšŠπšžπšπšŽ_𝚌𝚘𝚍𝚎. to πš™πš’. So both agents land on the same panels and stay comparable. Same repo, same week, one axis. github.com/desek/agent-obser…
6
1
2
206
Your agent now remembers the approaches you rejected, not just the code that survived. Nothing in a repo records what was ruled out. Code shows what worked, and squash merge deletes the commit bodies where the reasoning lived. So the agent confidently retries the approach you threw away six weeks ago. Shipped two skills that close the loop: β†’ /πšŒπš‘πšŽπšŒπš”πš™πš˜πš’πš—πš-πš’πšπšŽπš›πšŠπšπšŽ keeps a ledger of every attempt, including the discarded ones β†’ /πšŒπš‘πšŽπšŒπš”πš™πš˜πš’πš—πš-πšπš’πšœπšπš’πš•πš• ranks which of them deserve to be a standing rule β†’ nothing lands in π™°π™Άπ™΄π™½πšƒπš‚.πš–πš until you approve it Rules expire on purpose. A workaround for someone else's bug has to carry the test that retires it, so it dies with the bug instead of outliving it. npx skills add desek/governance
29
I built an MCP server that lets you manage your Outlook calendar and email from any AI assistant. Single Go binary. Runs locally. No cloud relay. No Entra ID setup. Just `go install` and start talking to your calendar. github.com/desek/outlook-loc… #MCP #AItools #AgenticAI
58
Daniel Grenemark retweeted
I prompted ChatGPT to build a futures trading model using Python and a Neural Net, integrated with market data from Sierra Chart. The model provides entries, exits and stop levels. ChatGPT is up 46% this week trading ES and ZN. Reply and Retweet and I'll DM the model.
621
530
37
706
131,129
Winning the hot sauce competition for sure. It was a true pleasure hanging around at Dreamhack with you @BillyPacman πŸ•Ή
2
1
If only the @Deezer desktop app could (be scary and) adapt my Flow selection based on analyzing my app/window title usage. Would love if it noticed how I switch from coding in @code to fanatic googling in @googlechrome and adapted the music selection to it.
1
Daniel Grenemark retweeted
Frequently get asked what monitoring tools someone should use to monitor their applications in #Azure. We published a new doc which helps explain what to use depending on your scenario. Check it out. Link: docs.microsoft.com/en-us/azu… #Azure #AzureMonitoring
3
68
1
86
Daniel Grenemark retweeted
It's been a huge piece of work, but it's done: here's more than half a billion passwords for you to download for free and use to help protect your systems. Or use the online k-Anonymity API developed in conjunction with @Cloudflare. It rocks! troyhunt.com/ive-just-launch…
46
1,074
63
1,775
Daniel Grenemark retweeted
ICYMI: Presentation Translator that lets you add live subtitles to your presentations in PowerPoint! #AI #Azure microsoft.com/en-us/download…
14
12
Daniel Grenemark retweeted
1
19
34
If you want to learn more about Docker, Kubernetes, Git & other modern tech, there is no better place than katacoda.com/ @Ben_Hall
3
18
1
39
Anyone got a good whitepaper for managing scalabilty of @nodejs (with or without @Docker) solutions? #DevOps
1
Just did hands-on competitive/usability re-pro of @googlecloud against my deep @Azure knowledge and recent @awscloud gig #mindblown #wtf
1
1
2